Evaluating biological plausibility of learning algorithms the lazy way

Owen Marschall, Kyunghyun Cho, Cristina Savin

TL;DR: We evaluate new ML learning algorithms' biological plausibility in the abstract based on mathematical operations needed

Keywords: Machine learning, back propagation through time, biological plausibility, online learning

Abstract: To which extent can successful machine learning inform our understanding of biological learning? One popular avenue of inquiry in recent years has been to directly map such algorithms into a realistic circuit implementation. Here we focus on learning in recurrent networks and investigate a range of learning algorithms. Our approach decomposes them into their computational building blocks and discusses their abstract potential as biological operations. This alternative strategy provides a “lazy” but principled way of evaluating ML ideas in terms of their biological plausibility